Inheritance diagram for q3::execution::gpu::lapack::SYCLGPUEigensolver:

Public Member Functions
	SYCLGPUEigensolver (std::shared_ptr< device::GPUDeviceBackend > device_backend)

	~SYCLGPUEigensolver () override=default

void	generalized_hermitian_eigenvalue_solve (int matrix_dimension, Complex matrix_a_data, int leading_dimension_a, Complex matrix_b_data, int leading_dimension_b, Real *eigenvalues) override

void	hermitian_eigenvalue_solve (int matrix_dimension, Complex matrix_data, int leading_dimension, Real eigenvalues) override

Public Member Functions inherited from q3::execution::gpu::lapack::GPUEigensolver
	GPUEigensolver (std::shared_ptr< device::GPUDeviceBackend > device_backend)

virtual	~GPUEigensolver ()=default

device::GPUDeviceBackend &	device_backend ()

const device::GPUDeviceBackend &	device_backend () const

Additional Inherited Members
Protected Attributes inherited from q3::execution::gpu::lapack::GPUEigensolver
std::shared_ptr< device::GPUDeviceBackend >	device_backend_

Detailed Description

Definition at line 24 of file sycl.cpp.

Constructor & Destructor Documentation

◆ SYCLGPUEigensolver()

q3::execution::gpu::lapack::SYCLGPUEigensolver::SYCLGPUEigensolver ( std::shared_ptr< device::GPUDeviceBackend > device_backend )

inlineexplicit

Construct a oneMKL LAPACK eigensolver.

Obtains the SYCL queue from the device backend via dynamic_cast. Throws if the device backend is not a SYCLGPUDeviceBackend.

Definition at line 32 of file sycl.cpp.

References q3::execution::gpu::lapack::GPUEigensolver::device_backend_.

◆ ~SYCLGPUEigensolver()

q3::execution::gpu::lapack::SYCLGPUEigensolver::~SYCLGPUEigensolver ( )

overridedefault

Member Function Documentation

◆ generalized_hermitian_eigenvalue_solve()

void q3::execution::gpu::lapack::SYCLGPUEigensolver::generalized_hermitian_eigenvalue_solve	(	int	matrix_dimension,
		Complex *	matrix_a_data,
		int	leading_dimension_a,
		Complex *	matrix_b_data,
		int	leading_dimension_b,
		Real *	eigenvalues
	)

inlineoverridevirtual

Solve A · x = λ · B · x via oneMKL hegvd.

B must be positive definite. On exit, matrix_a_data holds the eigenvectors and eigenvalues contains eigenvalues in ascending order.

Implements q3::execution::gpu::lapack::GPUEigensolver.

Definition at line 101 of file sycl.cpp.

◆ hermitian_eigenvalue_solve()

void q3::execution::gpu::lapack::SYCLGPUEigensolver::hermitian_eigenvalue_solve	(	int	matrix_dimension,
		Complex *	matrix_data,
		int	leading_dimension,
		Real *	eigenvalues
	)

inlineoverridevirtual

Solve A · x = λ · x via oneMKL heevd.

Uses the divide-and-conquer algorithm for better performance on larger matrices. On entry, matrix_data holds the Hermitian matrix (upper triangle). On exit, it holds the eigenvectors.

Implements q3::execution::gpu::lapack::GPUEigensolver.

Definition at line 51 of file sycl.cpp.

The documentation for this class was generated from the following file:

src/layer_0_core/math/linear_algebra/gpu/lapack/sycl.cpp

Public Member Functions